Which kind of proves my point... PROFITS are what is keeping tech advancement back - NOT actual scientific innovation.
If something isn't financially viable then what incentive is there for companies to pursue it?
Both AMD and Nvidia are constrained by the fabrication process. Both IHV's only option has been to increase the GPU die package up to the present limits of manufacturing.
TSMC is the only present option for performance GPUs (Intel aside). Every other foundry left the market because of the technical hurdles imposed and financial viability nearly a decade ago, with the exception of UMC and Chartered acting as second source for TSMC on rare occasions.
TSMC, as with every other foundry, has technical hurdles to overcome with smaller process nodes- both lithography tool availability (
ASML has a backlog stretching years), and the fact that transistor viability is still a stumbling block in
commercial production.
We're not (at least, I'm not) making judgement calls on whether it's right or wrong... simply that Nvidia (and possibly AMD) COULD be making better cards... of course, it isn't financially beneficial to them... so the consumers suffer!
And just where do you propose they build these GPUs? Stark Industries isn't a real company.
In what world does jamming every feature into a GPU bloating its size, blowing its power budget, and abrogating the ATX spec make even the slightest amount of sense when pleasing a fraction of the gaming fraternity simultaneously alienating OEMs and destroying markets where performance needs to balanced against power draw.
By all means tilt at windmills, but in the real world, R&D needs to be balanced by practicality - and while the enthusiast in us is disappointed that
practicality trumps endeavour, what you are advocating is for companies to emulate Xerox's PARC - R&D and bringing projects to fruition without financial consideration, as
this brief article outlines.
“There is a wide difference between completing an invention and putting the manufactured article on the market.” — Thomas Alva Edison
As for scaling - I once again would like to see your proof...Yes, AMD USED to rule the scaling world - up to around 2011....
On balance, when it works, Crossfire scaling is better (from a raw f.p.s. PoV) than SLI at the present time with the implementation of XDMA. That isn't an opinion but fact.
I wouldn't hold up the quadfire 295X2 as indicative of overall performance simply because the cards suffer issues, not least of which is CPU/system memory limitation, and poor driver support.
Quad multi-GPU suffers with both vendors as a general rule.
Firstly the driver shortcomings are BS. I build both brand machines and have no more problems with AMD than do I Nvidia.
Secondly your numbers are off, it is not 10%. I don't care about your link, I do my own partly for that reason as well as simply building high end machines. the majority of that figure comes from a very few games that are not AMD friendly. if I was so inclined I could put up a game lineup that would have the AMD's up by 10% as well, however I do not as I actually bench for accuracy. BY the way, why not TS's review of the Fury?....because you don't like the very close results. PC gamer is not gospel, so don't link me there like you just empirically proved anything.
I'm inclined to agree...mostly. Depending on the benchmark suite you can make a case for either card (Fury X or 980 Ti).
But for me personally as a tinkerer, the 980 Ti gets the nod. Fury X's overclocking is limited at present aside from third party tools and what seems a bug in CCC. AMD advertised the card as an overclockers dream yet voltage locked the card. That was done for a reason, and I suspect that reason is adverse effects - power draw, needing to ramp that Gentle Typhoon fan past the 1500 rpm and into noisy territory, localized heat generation and the PCB acting as an unwelcome heatsink (as per the 295X2).
Nearly every AIB custom 980 Ti clocks to 1500-1550 core / 8000-8400 effective memory on air, with at least one vendor (GALAX) offing a fully voltage unlocked card, with the probability that EVGA's Classy KPE will do the same. GM 200 I think has more in reserve than Fiji, and for me that offsets the AIO and promises of better drivers sometime in the future ( I suspect that AMD's driver team are fully engaged with Win10 at the moment).